Short answer: ultracode is a Claude Code setting, not an effort level. With it on, Claude plans a dynamic workflow for each real task: a script that runs many subagents in parallel, often dozens, and has them check each other's work before you see the result. Turn it on with /effort ultracode for the session, or type the word ultracode in a single prompt to run just that task as a workflow. It gives the most thorough results Claude Code can, and uses far more of your plan.
Everything here is from Anthropic's Claude Code documentation, checked on November 9, 2026. (I build Morphy, whose Claude chats offer ultracode too; that's near the end.)
What ultracode does
Normally Claude works through a task turn by turn, perhaps handing a few parts to subagents or a team (see subagents vs agent teams). With ultracode on, Claude decides when a task deserves a workflow, writes a JavaScript script for it, and a runtime runs that script in the background while your session stays free. One request can become several workflows in a row: one to understand the code, one to make the change, one to verify it.
What that buys you, in Anthropic's words, is a more trustworthy result than a single pass: independent agents can adversarially review each other's findings, or draft a plan from several angles and weigh them. Anthropic's own example from launch: Bun's port from Zig to Rust, 750,000 lines in eleven days, with 99.8% of the test suite passing.
| Subagents | Agent teams | Workflows (ultracode) | |
|---|---|---|---|
| Who decides what runs next | Claude, turn by turn | A lead agent | The script |
| Scale | A few tasks per turn | A handful of long-running peers | Dozens to hundreds of agents per run |
| If interrupted | Restarts the turn | Teammates keep running | Resumes in the same session |
How to turn ultracode on
| How | What it does |
|---|---|
ultracode anywhere in a prompt | Runs that one task as a workflow, without changing the session |
/effort ultracode | Turns it on for the rest of the session |
claude --effort ultracode | Starts a session with it on, at xhigh effort |
"ultracode": true in settings | Every session starts with it on |
/effort, then Tab | Flips the Ultracode toggle in the effort slider |
/effort ultracode off | Turns it off again |
Asking in your own words ("use a workflow") works like the keyword. If you typed ultracode without meaning it, Alt+W (Option+W on a Mac) dismisses it for that prompt, and Ultracode keyword trigger in /config turns the keyword off. The keyword only counts in a prompt you type yourself: not with claude -p, in scheduled tasks or in a pull request comment.
/effort ultracode off, the slider toggle, and ultracode at other effort levels than xhigh need Claude Code 2.1.284 or later; run claude update if /effort ultracode off says Invalid argument.
Ultracode vs max effort
They do different things, and they combine.
max effort | ultracode | |
|---|---|---|
| What it is | The deepest reasoning level | A setting that fans work out to many agents |
| What changes | Claude thinks longer before each answer | Claude writes and runs a workflow for each real task |
| Good for | One hard problem, like a subtle bug | Big jobs: audits, migrations, research |
| Downside | Diminishing returns, prone to overthinking | Much more usage per request |
| How to keep it on | CLAUDE_CODE_EFFORT_LEVEL=max (settings don't accept max) | The ultracode setting |
Ultracode runs at whatever effort the session is on, so /effort max with ultracode on gives you both. Opus 5.5, Sonnet 5.5 and Haiku 5.5 default to medium effort; claude --effort ultracode starts at xhigh. More on choosing a level in Anthropic's model configuration guide.
What ultracode costs
Each request uses more tokens and takes longer than without a workflow. On a Claude plan those tokens come out of your 5-hour and weekly limits, so a session with ultracode on reaches them sooner. Anthropic's advice: turn it off when you're back to routine work.
With ultracode on, Claude Code also drops three of its brakes, because turning it on already opts you in to large runs:
- the Large workflow warning (normally at more than 25 agents or 1.5 million tokens),
- the session's limit on how many subagents run at once,
- in auto mode, the approval before the first workflow.
What still holds: at most 16 agents at a time by default (fewer on a computer with fewer cores) and 1,000 agents per run. To keep runs smaller, set the Dynamic workflow size in /config: small (fewer than 5 agents), medium (fewer than 10), large (fewer than 50) or unrestricted. The default is medium, and small on Pro. You can watch each agent's token use in /workflows and stop a run there.
If a run hits your plan's limit, it waits for the reset and carries on by itself (Claude Code 2.1.271 or later, in an interactive session). On plans with a 5-hour limit, that wait is the usual brake: Claude's Ultra plan has none for verified accounts, which is why it suits long workflows.
Which plans have it?
Dynamic workflows, and with them ultracode, are on every paid Claude plan (Pro, Max, Ultra, Team and Enterprise), with an Anthropic API key, and on Amazon Bedrock, Google Cloud and Microsoft Foundry. On Pro, turn on Dynamic workflows in /config first. They work in the terminal, the desktop app and the IDE extensions.
Ultracode isn't offered when workflows are off, or when the model doesn't support xhigh effort; --effort ultracode then starts at the highest level the model allows.
Try one without ultracode
Claude Code comes with one workflow built in: /deep-research, which searches the web from several angles, cross-checks the sources and returns one cited report.
/deep-research What changed in the Node.js permission model between v20 and v22?Run /workflows to watch it: each phase with its agents, and what each one found. p pauses, x stops, and s saves a run's script as a command of your own, in .claude/workflows/ (shared with the repository) or ~/.claude/workflows/ (only you).
Turn workflows off
- For yourself: Dynamic workflows in
/config,"disableWorkflows": truein~/.claude/settings.json, orCLAUDE_CODE_DISABLE_WORKFLOWS=1. - For an organization:
"disableWorkflows": truein managed settings, or the toggle in Claude Code's admin settings.
With workflows off, the keyword does nothing and the Ultracode toggle disappears from /effort.
In Morphy
In a Morphy chat on Claude, ultracode is the last choice under Effort in the model menu: it runs at Extra high, and for every real task Claude plans and runs a team of agents instead of working alone. It needs Claude Code's workflows on your Claude plan; without them, the chat runs at Extra high. Because it uses so much more of your plan, Settings → Plan usage is worth a look: it shows how much of your Claude plan is left and when it resets. See Models and effort.
Questions
Is ultracode an effort level? No. It's a Claude Code setting that runs dynamic workflows at whichever effort level the session is on. Starting with claude --effort ultracode sets that level to xhigh.
What's the difference between ultracode and max? max makes Claude think longer before each answer. Ultracode makes Claude split real tasks across many agents that check each other's work. You can use both at once.
Does ultracode use more of my plan? Yes, much more: every request runs a workflow of several agents, and all their tokens count toward your 5-hour and weekly limits.
Is ultracode available on Claude Pro? Yes, once you turn on Dynamic workflows in /config. On Pro, workflows aim for fewer than 5 agents unless you change the size.
How do I turn ultracode off? /effort ultracode off, or remove "ultracode": true from your settings if you set it there.
Related guides
- Claude Max vs Ultra: which plan to pick
- How to stop hitting Claude Code's usage limits
- How to run multiple Claude Code agents at once
- Claude Code worktrees
References
- Anthropic: dynamic workflows, effort levels and ultracode, introducing dynamic workflows (May 28, 2026)
- Morphy: Models and effort, Plan usage
